JPH10228461A - Demand prediction method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To predict a demand in a stage where the products of a short life cycle cannot sufficiently be predicted only with the result data. SOLUTION: A parameter estimation function 10 allocates the shift of the sales amount result to a growth model expressed by an accumulated sales amount Nt counted from sales start to time (t) against the product where sales result data 30 of the whole life cycle is obtained, and parameters (p), (q) and (m) adapted to the product are estimated and they are preserved in an estimation parameter 40. The demand prediction function 20 selects the appropriate estimation value for the product being the demand prediction object during sales among the estimation values and substitutes the parameter for the expression so as to predict demand.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a demand forecasting method based on past performance data, and more particularly to a demand forecasting method for a product having a short life cycle from the start of sales to the end of sales.

[0002]

2. Description of the Related Art In the case of products that are greatly influenced by fashion and changes in technology, such as clothes and personal computers, the life cycle from the start of sales to the end of sales of a certain product is as follows.
A few months to a year, short. A growth model proposed by Bass is known as one of the models representing the demand pattern of such a product. This is a model that expresses a so-called mountain-shaped demand pattern in which demand rapidly increases immediately after the start of sales and monotonously decreases after the peak, and the cumulative demand Nt at time t is Next formula

[0003]

(Equation 1)

[0004] Here, m, p, and q are parameters that characterize the demand pattern, m is the total demand, and p and q determine the demand curve.

On the other hand, Matsuo et al. Have proposed a model in which seasonal variation factors are considered in the above model. In this model, the cumulative demand Nt at time t is given by

[0006]

(Equation 3)

Is represented by Here, αt is a seasonal variation factor and generally represents a monthly characteristic, and αt = αt−12t> 12.

Regarding the above demand model, the following paper
Abbas A. Kurawarwala and Hirofumi Matsuo: Product
growth models for medium term forecasting of short
life cycle products.Technical Report 92-09-04, I
C2 Institute, The University of Texas at Austin, 1
This is described in detail in 992.

[0009]

If the product has a long life cycle to some extent, the model under consideration is fitted to past performance values, parameters are estimated,
It is possible to perform demand prediction beyond that. However, for our short life cycle products, we are already past the product's peak sales when we have enough performance data to estimate the parameters, at which point we anticipate future demand. However, there is a problem that it cannot be used for parts procurement planning and production planning.

The present invention has been made in view of such circumstances, and an object of the present invention is to provide a product having a short life cycle at a stage where it is possible to provide feedback to a production plan immediately after the start of sales or shipment. And a means for performing demand forecasting.

[0011]

In order to solve the above-mentioned problems, according to the present invention, first, means for periodically recording the actual sales volume of each product, and the above-mentioned means from the start of sales to the end of sales. For products for which the actual sales volume has been recorded, the change in the actual sales volume is calculated by calculating

[0012]

(Equation 1)

The means for estimating the parameters p, q, and m applicable to the product by applying the growth model expressed by the above, and associating the estimated p, q, and m with the recorded sales volume results A means for storing is provided. The above processing is performed for a plurality of products, data is accumulated, and used for demand prediction.

[0014] In the first aspect, if a product whose demand is to be predicted has elapsed from the start of sales by the time t, the actual sales volume up to that time and the saved past Comparing the actual sales volume from the start of sales to the point in time t, and comparing the products with the closest cumulative sales volume from the start of sales to the point in time t with the patterns of the actual sales volume from the start of sales to the point in time t Means to select the closest product and the estimated value m for the product whose cumulative sales volume is the closest
And, means for estimating the estimated value p, q for the product with the closest sales volume actual pattern as the parameters p, q, m of the product for sale, using the estimated p, q, m Then, the subsequent demand amount of the product for sale, which is being sold, is predicted. This makes it possible to predict the demand for the product immediately after the start of sales.

From a second viewpoint, p and q are expressed as follows:
As in the first aspect, the product with the closest sales performance pattern is selected and the estimated value of the product is used, and the parameter m is calculated as follows: Cumulative sales volume m1 from the start of product sales until time t, cumulative sales volume m2 until sales stop
Using the following equation

[0016]

(Equation 2)

[0017]

In the third aspect, as for the m, as in the first aspect, a product having the closest cumulative demand from the start of sales to time t is selected, and the estimated value of the product is used. , Parameters p and q are calculated by adding

[0019]

(Equation 1)

The growth model is applied to estimate p and q.

On the other hand, in the present invention, as a demand forecasting model, the cumulative sales amount Nit from the start of sales of the product i to the time t is calculated by the following equation.

[0022]

(Equation 3)

Also provided is one using the growth model expressed by:

That is, the model is applied to the means for periodically recording the actual sales volume for each product, and the product for which the actual sales volume is recorded from the start of sales to the end of sales. Means for estimating the parameters pi, qi, mi and the parameter αt common to all products, means for storing the estimated pi, qi, mi and the recorded sales volume results in association with each other, and means for storing αt Is provided. Further, for the demand forecast target product, as in any of the first to third aspects described above, pi, qi, and mi are estimated, and the stored αt is used to sell the Forecast the subsequent demand for the forecasted product.

From another viewpoint, regarding data used for sales volume comparison, a product obtained by dividing the actual value by αt as the sales volume of each product in t months is used.

Alternatively, the cumulative sales volume from the start of sales of the past product i to time t is expressed by the following equation.

[0027]

(Equation 1)

Then, the values obtained by substituting the estimated values pi, qi, and mi corresponding to the product are used, and the product of the actual value divided by αt is used as the sales volume in t months of the product to be estimated. As a result, comparison can be performed with data from which the seasonal variation factor has been removed, so that more accurate estimation is possible.

From another viewpoint, when the actual sales volume from the start of sales to the end of sales of a certain product is recorded, means for estimating pi, qi, mi and αit from only the actual sales volume of the product are provided. And the distance between product i and product j is

[0030]

(Equation 4)

Each product is analyzed in a cluster,
In addition to performing grouping, it has means for determining a representative value of αt for each group, and when estimating a parameter of a product for sale or estimating demand, selecting one group from the above groups and selecting a group Use αt. Thus, for example, even when the same kind of product has different seasonal factors due to a detailed level of specification or a difference in sales channel, it is possible to perform a detailed demand forecast.

By the way, when applying the growth model with seasonal fluctuation factors, if αt is given every month, 15 parameters will be estimated. Therefore, as described from the above viewpoint, when applying a model independently for each product, there is a possibility that sufficient data may not be available due to a short life cycle. Therefore, from another viewpoint, there are means for estimating only pi, qi, and mi by fixing αit, and means for estimating αit by fixing pi, qi, and mi. As a first step, a process of estimating pi, qi, and mi is repeatedly performed by the above two estimating means until termination at an appropriate time, and pi, qi, m
Estimate i, αit. As a result, it is possible to apply a growth model to a product having a small number of performance data.

[0033]

Embodiments of the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited by this.

FIG. 1 is a functional information related diagram of the demand forecasting method according to the present invention. The present invention mainly includes two functions, a parameter estimation function 10 and a demand prediction function 20, and two pieces of information, sales performance data 30, and estimation parameter data 40. FIG. 2 shows a hardware configuration diagram for realizing this. The two functions in FIG. 1 are processed by the computing device 2 in FIG. The sales result data 30 is taken in from the data input device 1. And sales performance data 3
0, the estimation parameter data 40 is stored in the storage device 3. As the data input device 1, for example, a keyboard or a barcode reader is used. The computing device 2 may be a general-purpose computer. The storage device 3 is not particularly limited, such as a magnetic disk, a magnetic tape, and an optical disk. The output device 4 outputs a prediction result, and a display or a printer may be used, or the storage device 3 may be used instead. Further, the input device 1, the computing device 2, the storage device 3,
The output device 4 may be connected by a dedicated line or may be connected on a network via a LAN or the like.

Hereinafter, the processing flow of the two functions of FIG. 1 will be described. First, the parameter estimation function 10 will be described. This function is performed once for each product after the product has completed its life cycle and sales have ended. First, step 10
At 0, the sales result data 30 of the target product is read. In this embodiment, sales performance data totaled monthly is used.
It is also possible to use data tabulated weekly or daily. There is no restriction on the timing of data collection. Input device 1
When a keyboard is used, the sales amount of each product is input once a month and stored in the storage device 3, and during this processing, data over the entire life cycle may be read from the storage device 3, All data may be collectively input from the keyboard during this processing. Also, when using a barcode reader as an input device, each time a product is sold, the information is read using a barcode reader,
The read data is stored in the storage device 3 or another auxiliary storage device, and is totaled for each month before the start of this processing, as shown in FIG.

Next, in step 110, the cumulative sales volume Nt after t months is added to the sales performance data of the product by the following equation.

[0037]

(Equation 1)

The model represented by is applied. Specifically, the least squares estimation is calculated. That is, the target product i
Xt is the sales result data after t months from the start of sales of

[0039]

(Equation 5)

Here, si and li calculate p, q, and m, respectively, which minimize the function f1 represented by the first month in which the product i is sold and the last month in which the sales are recorded.

In step 120, p, q, and m obtained by the above calculation are used as the estimation parameters 40 in the storage device 3.
To save. At this time, the sales performance data 30 used in the above calculation is easily traceable from any information.
Set a bidirectional pointer.

Next, the processing flow of the demand prediction function 20 will be described.

First, in step 200, the sales performance data 30 is read for the product i for which demand is to be predicted. This reading method is the same as the method described in the parameter estimation function 10. However, here, data on the entire life cycle of the target product is not required. Rather, the data is not available for the entire life cycle, as the product is still being sold.
Now, suppose that the performance data of the product i has been obtained only for a month from the start of sales.

Next, at step 210, based on the sales performance data 30, a product whose sales performance pattern from the start of sales to a month is closest to the product i to be predicted is selected. Specifically, the slope of the actual value of each month is compared, and a search is made for the one with the minimum squared difference. In other words, for any product j that has performance data for the entire life cycle,

[0045]

(Equation 6)

However, xi0 = xj0 = 0 is calculated, and j having the smallest value g is obtained.

Then, in step 220, the estimated values p and q for the product j are read from the estimated parameter data 40.

Next, in step 230, based on the sales performance data 30, a product whose cumulative sales volume from the start of sales to a month is closest to the product i to be predicted is selected. In particular,
For any product j with actual life cycle performance data

[0049]

(Equation 7)

Is calculated to determine j having the smallest value h.

Then, in step 240, the estimated value m for the product j is read from the estimated parameter data 40.

Then, in step 250, the above-mentioned p, q, and m are substituted into the above (Equation 1), and Nt is calculated for t> a. At this time, Nt−Nt−1 is the predicted value of the sales amount t months after the start of sales. In step 260, the result is output to the output device 4.

With the above-described processing, it is possible to predict the subsequent sales volume of the product on sale.

In the above embodiment, the demand forecasting function 20 searches for products whose sales performance pattern and cumulative sales volume are closest to the product to be forecast. However, two other methods will be described below.

FIG. 3 shows a processing flow of the demand forecasting function 21 of the second embodiment. This is the same as the demand forecasting function 20 of the first embodiment except for the method of determining m. Steps 230, 24
Steps 270 and 280 are executed instead of 0. In step 270, the sales performance data 30 of the product obtained in step 210 is read, and the cumulative sales volume m1 from the start of sales to a month and the cumulative sales volume m2 at the end point are calculated. Further, the cumulative sales volume m0 from the start of sales of the demand forecast target product to a month is calculated.

Then, in step 280, the following equation

[0057]

(Equation 2)

Is calculated to obtain m.

As described above, in the second embodiment, prediction is performed only from information on products having similar sales performance patterns.

FIG. 4 shows a processing flow of the demand forecasting function 22 of the third embodiment. This is the same as the demand forecasting function 20 of the first embodiment except for the method of determining p and q. First, step 23
According to 0 and 240, m is determined. Next, step 29
At 0, a model in which m obtained above is substituted into (Equation 1) for the actual data of the demand prediction target product is applied. That is, p and q that minimize the function f ′ obtained by replacing li with a in (Equation 5) are obtained.

As described above, in the third embodiment, prediction is performed only from information on products having similar cumulative sales amounts.

In the embodiment described above, the demand forecasting model of (Equation 1) is used. However, the cumulative sales amount Nit from the start of the sales of the product i to the time t is expressed by the following equation.

[0063]

(Equation 3)

By using the growth model expressed by the following equation, the seasonal fluctuation factor can be considered. Here, the parameters pi, qi, and mi are specific to each product, and the parameter α
t is common to all products. Then, αt = αt−12t> 12.

Hereinafter, the embodiment will be described. FIG. 5 shows a function information related diagram of the present embodiment. Basically, the functions and information are the same as those shown in FIG. Only the differences are explained.

As information, in addition to the sales performance data 30 and the estimated parameter data 40, there is a seasonal factor parameter data 50. In this, the estimated αt (t = 1,..., 12) is stored.

In the parameter estimation function 11, first, in step 130, the sales performance data 30 is read for all products having sales performance data throughout the entire life cycle from the start of sales to the end of sales.

Next, in step 130, the model of (Equation 3) is applied to all the above products. In particular,
Next formula

[0069]

(Equation 8)

Here, n represents the total number of target products. Pi, qi, mi (i = 1,...,
Calculate n) and αt.

Further, in step 150, pi, qi, and mi obtained by the above calculation are used as the estimated values of the products corresponding to the estimation parameters 40, and αt obtained by the above calculation is used as the seasonal variation parameter 50, respectively. The data is stored in the storage device 3.

On the other hand, in the demand forecasting function 23, step 3
At 00, the seasonal variation parameter 50 stored above is read. Then, in step 310, p, q, and m determined in steps 220 and 240 and αt read in above are substituted for (Equation 3), and Nit is calculated for t> a. At this time, Nit-Ni, t-1 is a predicted value of the sales volume t months after the start of sales.

In the demand forecasting method described above, since seasonal fluctuation factors can be considered, more accurate forecasting becomes possible.

In the demand forecasting function 23 of the above embodiment, p, q, and p are calculated by the method described in the second and third embodiments.
m may be determined.

In the above embodiment, the demand forecasting function 23
In the above, the sales performance data 30 is used as it is, but data obtained by removing the seasonal fluctuation factor may be used. Specifically, prior to step 210, the following step 205 is executed.

Step 205: The value obtained by dividing the actual sales value of each product in j months by αj is regarded as the actual value and used for the following calculation.

Alternatively, the following step 206 is executed.

Step 206: For a product for which sales performance data exists over the entire life cycle and the parameter estimation function 11 is executed and the estimated values of p, q, and m are recorded in the estimated parameters 30, Nt obtained by substituting the estimated values pi, qi, and mi of the product into 1) is regarded as the actual value of the cumulative sales volume. For the demand forecast target product, the actual sales value in month j is divided by αj. The value is regarded as the actual value and used for the following calculation.

As a result, comparison can be performed using data from which seasonal fluctuation factors have been removed, so that more accurate prediction is possible.

In the embodiment described so far, the seasonal factor parameter αj is common to all products. Next, the products are divided into several groups, and the seasonal factor parameters are prepared for each group. In the demand prediction function, an embodiment in which one group is selected from the functions will be described.

First, in the estimation parameter data 41, an area for storing αj and the ID of the group to which the product belongs is extended in addition to p, q, and m for each product. Further, in the seasonal variation parameter 51, the area is extended so that a plurality of sets of αj can be stored together with each group ID. This is shown in FIG.
Shown in

FIG. 7 shows a processing flow of the parameter estimating function 12 of this embodiment. This process is performed once for a product whose life cycle has been completed.

First, in step 100, the sales result data 30 of the target product is read.

Next, in step 160, the model in which the cumulative sales volume Nit after t months is expressed by (Equation 3) is applied to the sales performance data of the product i, and pi, Estimate qi, mi, αit.

Then, in step 170, the estimated value is stored in the estimated parameter data 41.

Further, in step 180, each seasonal factor parameter αit is set for all the products estimated so far.
Cluster analysis is performed based on the estimated value of. Specifically, first, the distance between product i and product j is

[0087]

(Equation 4)

Is defined. If the distance is within a certain value, they are grouped into the same group.

Next, in step 190, for each of the above groups, the average of the seasonal factor parameter αit of the product belonging to the group is calculated for each t, and the average is taken as the representative value of the seasonal factor parameter of the group. Is stored in the seasonal factor parameter 51.

The demand forecasting function is basically the same as the processing flow shown in FIG. Only step 300
Then, processing for selecting one group is added. As a method of the selection, for example, a group to which a product whose type, sales channel, and the like are similar to the demand prediction target product belongs is selected.

Thus, for example, even when the same kind of product has different seasonal factors due to detailed levels of specifications and differences in sales channels, it is possible to perform fine demand prediction.

By the way, in step 160 in the above embodiment,
Therefore, when a model is applied independently for each product, there are 15 parameters to be estimated, but the life cycle is short and the number of data is less than that, so estimation may not be possible. The parameter estimation method in this case will be described below.

FIG. 8 shows the flow of processing in this embodiment.

First, in step 400, αit is all 1
Set to.

Next, in step 410, αit obtained above or obtained in step 420 is substituted for (Equation 3), and pi,
Estimate only qi and mi. At this time, the value of f2 when the estimated value is substituted into (Equation 8), that is, the minimum value of f2 is stored as r1.

Further, in step 420, pi, qi, and mi obtained in step 410 are substituted for (Equation 3), and
Estimate αit. Then, as in step 420, f2
Is stored as r2.

Then, in step 430, r1 and r2 are compared. If the reduction ratio (r1-r2) / r1 of r2 to r1 exceeds a certain reference value, the process returns to step 410. If not, the process is terminated, and pi, qi, mi obtained when step 410 is performed last and finally step 41
Αit obtained when 0 is performed is set as an estimated value.

As a result, it is possible to apply a growth model in consideration of seasonal fluctuation factors even for a product having a small number of actual data.

[0099]

According to the present invention, there is an effect that demand for a product having a short life cycle can be predicted.

[Brief description of the drawings]

FIG. 1 is a functional information related diagram according to a first embodiment of the present invention.

FIG. 2 is a hardware configuration diagram for realizing the present invention.

FIG. 3 is a flowchart of a process of a demand prediction function according to a second embodiment of the present invention.

FIG. 4 is a flowchart of a process of a demand prediction function according to a third embodiment of the present invention.

FIG. 5 is a function information related diagram according to a fourth embodiment of the present invention.

FIG. 6 is a diagram showing information used in a sixth embodiment of the present invention.

FIG. 7 is a process flow of a parameter estimation function according to a sixth embodiment of the present invention.

FIG. 8 is a flowchart showing a procedure of parameter estimation according to a seventh embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input device, 2 ... Calculation device, 3 ... Storage device, 4 ... Output device, 10 ... Parameter estimation function, 20 ... Demand forecasting function, 30 ... Sales performance data, 40 ... Estimation parameter data, 50 ... Seasonal factor parameter data

Claims (12)

[Claims] 1. Means for periodically recording the actual sales volume for each product, and for a product on which the actual sales volume is recorded from the start of sales to the end of sales, the transition of the actual sales volume is started. The cumulative sales volume Nt up to the time t calculated from the following formula is expressed by the following equation. Means for estimating the parameters p, q, m adapted to the product by applying to the growth model expressed by, and means for associating the estimated p, q, m with the recorded sales results and storing At a certain point in time, the actual sales volume of a product on sale that has elapsed from the start of sales by the time t and the sales volume up to the time t from the start of sales of the stored past product A means for comparing the actual performance and selecting a product having the closest cumulative sales volume from the start of sales to time t and a product having the closest sales volume performance pattern from the start of sales to time t, Means for estimating the estimated value m for the closest product and the estimated values p and q for the product whose sales volume actual pattern is the closest as the parameters p, q and m of the sold product, and On sale using p, q, m A demand forecasting method characterized by forecasting a subsequent demand quantity of a product. 2. In the parameter estimation of a product for sale according to the demand forecasting method according to claim 1, for parameters p and q, a product having the closest sales performance pattern is selected, and an estimated value of the product is used. For the parameter m, using the cumulative sales volume m1 of the estimation target product up to the present time, the cumulative sales volume m2 from the start of sales of the selected product to the time t, and the cumulative sales volume m3 until the sales stop, The following equation A demand forecasting method characterized by the following: 3. In the parameter estimation of a product for sale according to the demand forecasting method according to claim 1, the parameter m
The product with the closest cumulative demand up to time t from the start of sales is selected, the estimated value of the product is used, and the parameters p and q are added to the actual data of the product to be estimated by the following equation. Substituting m obtained above into the growth model,
A demand forecasting method characterized by estimating p and q. 4. A means for recording the monthly sales volume of each product, and a change in the sales volume of the product i for all the products in which the sales volume is recorded from the start of the sales to the end of the sales. Is the cumulative sales volume Nit up to t months from the start of sales
Is given by However, si represents the sales start month of the product i and is applied to the growth model expressed, and the above-mentioned product-specific parameters pi, qi, mi and the parameter αt (t = 1, 2, ... .)
Means for estimating the above, pi, qi, mi and the means for storing the recorded sales amount in association with each other, and the means for storing the estimated αt, at a certain time, from the start of sales Product i0 sold for at least t months
Compare the actual sales volume and the actual sales volume up to time t from the start of sales of the stored past products above, and compare the product with the closest cumulative sales volume up to time t from the start of sales. i1 and means for selecting a product i2 having the closest sales volume pattern until t months from the start of sales, and an estimated value mi1 for the product i1 having the closest cumulative sales volume,
The estimated values pi2 and qi2 for the product i2 having the closest sales amount actual pattern are calculated by using the parameters pi0 and q of the product i0 on sale.
having means for estimating i0 and mi0,
A demand forecasting method characterized by forecasting a subsequent demand quantity of a product for sale using 0, qi0, mi0 and the stored αt. 5. In the parameter estimation of a product for sale according to the demand forecasting method according to claim 4, for the parameters p and q, a product having the closest sales performance pattern is selected, and the estimated value of the product is used. Regarding the parameter m, using the cumulative sales volume m0 of the target product to be estimated up to the present time, the cumulative sales volume m1 up to t months from the start of sales of the selected product, and the cumulative sales volume m2 until sales stop, The following equation A demand forecasting method characterized by the following: 6. The parameter estimation of a product for sale according to the demand forecasting method according to claim 4, wherein the parameter m is
The product with the closest cumulative demand up to time t from the start of sales is selected, the estimated value of the product is used, and the parameters p and q are added to the actual data of the product to be estimated by the following equation. Where si is the growth model in which the above obtained m and the saved αj are substituted to represent the sales start month of the product i, and p, q
A demand forecasting method characterized by estimating a demand. 7. With respect to data used for comparing sales amounts in parameter estimation of products for sale according to the demand forecasting method according to claim 4, the actual value is represented by αt as the sales amount of each product in t months.
A demand forecasting method characterized by using a value divided by 8. The data used for comparing the sales volume in the parameter estimation of the product for sale according to the demand forecasting method according to claim 4 is calculated as the cumulative sales volume up to t months from the start of sales of the past product i. Equation 1 Using the values obtained by substituting the estimated values pi, qi, and mi corresponding to the product in question, and using the actual value divided by αt as the sales volume of the target product in t months. Demand forecasting method. 9. The demand forecasting method according to claim 4, wherein when the actual sales volume from the start of sales to the end of sales of a certain product is recorded, pi, q
The means for estimating i, mi, αit and the distance between product i and product j are given by Has a means for performing a cluster analysis of each product and performing grouping, and determining a representative value of αt for each group. A demand forecasting method, wherein a group is selected from the group and αt of the group is used. 10. The demand forecasting method according to claim 9, wherein α
a means of fixing it and estimating only pi, qi, mi, and pi, q
a means for estimating αit by fixing i, mi, and estimating pi, qi, mi by fixing αit to all 1s as a first step. A demand forecasting method characterized by repeating means, and estimating pi, qi, mi, and αit. 11. The demand forecasting method according to claim 1, wherein actual shipment volume is used instead of actual sales volume. 12. The demand forecasting system according to claim 4, wherein the actual sales volume is recorded every week, and the change in the actual sales volume of the product i is counted from the start of sales to the cumulative sales volume until t weeks. Nit is given by However, si is a demand forecasting method characterized by using a growth model represented by a week indicating the sales start of product i.